Solving Constraint Satisfaction Problems (CSPs) using Search CPSC 322 – CSP 2 Textbook Poole and Mackworth: Sections §4.3-4.4 Lecturer: Alan Mackworth October 1, 2012

Lecture Overview Constraint Satisfaction Problems (CSPs): Definition and Recap CSPs: Motivation Solving CSPs Generate & Test Graph search

Constraint Satisfaction Course Overview Course Module Representation Environment Reasoning Technique Deterministic Stochastic Problem Type Arc Consistency Constraint Satisfaction Variables + Constraints Search Static Bayesian Networks Logics Logic Search Variable Elimination Uncertainty Decision Networks Sequential STRIPS Variable Elimination Planning Planning Decision Theory Search Now focus on CSPs Markov Processes Value Iteration

Standard Search vs. CSP First studied general state space search in isolation Standard search problem: search in a state space State is a “black box” - any arbitrary data structure that supports three problem-specific routines: goal test: goal(state) finding successor nodes: neighbors(state) if applicable, heuristic evaluation function: h(state) We’ll see more specialized versions of search for various problems

Search in Specific R&R Systems Constraint Satisfaction Problems: State Successor function Goal test Solution Heuristic function Planning : Inference

Constraint Satisfaction Problems (CSPs): Definition Definition: A constraint satisfaction problem (CSP) consists of: a set of variables V a domain dom(V) for each variable V V a set of constraints C Simple example: V = {V1} dom(V1) = {1,2,3,4} C = {C1,C2} C1: V1  2 C2: V1 > 1 Another example: V = {V1,V2} dom(V1) = {1,2,3} dom(V2) = {1,2} C = {C1,C2,C3} C1: V2  2 C2: V1 + V2 < 5 C3: V1 > V2

Constraint Satisfaction Problems (CSPs): Definition Definition: A constraint satisfaction problem (CSP) consists of: a set of variables V a domain dom(V) for each variable V V a set of constraints C Definition: A model of a CSP is an assignment of values to all of its variables that satisfies all of its constraints. Simple example: V = {V1} dom(V1) = {1,2,3,4} C = {C1,C2} C1: V1  2 C2: V1 > 1 All models for this CSP: {V1 = 3} {V1 = 4}

Constraint Satisfaction Problems (CSPs): Definition Definition: A constraint satisfaction problem (CSP) consists of: a set of variables V a domain dom(V) for each variable V V a set of constraints C Definition: A model of a CSP is an assignment of values to all of its variables that satisfies all of its constraints. Another example: V = {V1,V2} dom(V1) = {1,2,3} dom(V2) = {1,2} C = {C1,C2,C3} C1: V2  2 C2: V1 + V2 < 5 C3: V1 > V2 Which are models for this CSP? {V1=1, V2=1} {V1=2, V2=1} {V1=3, V2=1} {V1=3, V2=2}

Possible Worlds Definition: A possible world of a CSP is an assignment of values to all of its variables. i.e. a model is a possible world that satisfies all constraints Definition: A model of a CSP is an assignment of values to all of its variables that satisfies all of its constraints. Another example: V = {V1,V2} dom(V1) = {1,2,3} dom(V2) = {1,2} C = {C1,C2,C3} C1: V2  2 C2: V1 + V2 < 5 C3: V1 > V2 Possible worlds for this CSP: {V1=1, V2=1} {V1=1, V2=2} {V1=2, V2=1} (a model) {V1=2, V2=2} {V1=3, V2=1} (a model) {V1=3, V2=2}

Constraints Constraints are restrictions on the values that one or more variables can take Unary constraint: restriction involving a single variable E.g.: V2  2 k-ary constraint: restriction involving k different variables E.g. binary (k=2): V1 + V2 < 5 E.g. 3-ary: V1 + V2 + V4 < 5 We will mostly deal with binary constraints Constraints can be specified by listing all combinations of valid domain values for the variables participating in the constraint E.g. for constraint V1 > V2 and dom(V1) = {1,2,3} and dom(V2) = {1,2}: giving a function (predicate) that returns true if given values for each variable which satisfy the constraint else false: V1 > V2 V1 V2 2 1 3

Constraints A possible world satisfies a set of constraints if the values for the variables involved in each constraint are consistent with that constraint They are elements of the list of valid domain values Function returns true for those values Examples {V1=1, V2=1} (does not satisfy above constraint) {V1=3, V2=1} (satisfies above constraint) V1 V2 2 1 3

Scope of a constraint Definition: The scope of a constraint is the set of variables that are involved in the constraint Examples: V2  2 has scope {V2} V1 > V2 has scope {V1,V2} V1 + V2 + V4 < 5 has scope {V1,V2,V4} How many variables are in the scope of a k-ary constraint ? k variables

Finite Constraint Satisfaction Problem: Definition Definition: A finite constraint satisfaction problem (FCSP) is a CSP with a finite set of variables and a finite domain for each variable. We will only study finite CSPs here but many of the techniques carry over to countably infinite and continuous domains. We use CSP here to refer to FCSP. The scope of each constraint is automatically finite since it is a subset of the finite set of variables.

Examples: variables, domains, constraints Crossword Puzzle: variables are words that have to be filled in domains are English words of correct length (binary) constraints: words have the same letters at cells where they intersect Crossword 2: variables are cells (individual squares) domains are letters of the alphabet k-ary constraints: sequences of letters form valid English words (k= 2,3,4,5,6,7,8,9)

Examples: variables, domains, constraints Sudoku variables are cells domain of each variable is {1,2,3,4,5,6,7,8,9} constraints: rows, columns, boxes contain all different numbers How many possible worlds are there? (say, 53 empty cells) How many models are there in a typical Sudoku? 53*9 539 953 About 253 1 953

Examples: variables, domains, constraints Scheduling Problem: variables are different tasks that need to be scheduled (e.g., course in a university; job in a machine shop) domains are the different combinations of times and locations for each task (e.g., time/room for course; time/machine for job) constraints: tasks can't be scheduled in the same location at the same time; certain tasks can't be scheduled in different locations at the same time; some tasks must come earlier than others; etc. n-Queens problem variable: location of a queen on a chess board there are n of them in total, hence the name domains: grid coordinates constraints: no queen can attack another

Constraint Satisfaction Problems: Variants We may want to solve the following problems with a CSP: determine whether or not a model exists find a model find all of the models count the number of models find the best model, given some measure of model quality this is now an optimization problem determine whether some property of the variables holds in all models

Solving Constraint Satisfaction Problems Even the simplest problem of determining whether or not a model exists in a general CSP with finite domains is NP-hard There is no known algorithm with worst case polynomial runtime. We can't hope to find an algorithm that is polynomial for all CSPs. However, we can try to: find efficient (polynomial) consistency algorithms that reduce the size of the search space identify special cases for which algorithms are efficient work on approximation algorithms that can find good solutions quickly, even though they may offer no theoretical guarantees find algorithms that are fast on typical (not worst case) cases

Lecture Overview Constraint Satisfaction Problems (CSPs): Definition and Recap Constraint Satisfaction Problems (CSPs): Motivation Solving Constraint Satisfaction Problems (CSPs) Generate & Test Graph search

CSP/logic: formal verification Hardware verification Software verification (e.g., IBM) (small to medium programs) Most progress in the last 10 years based on: Encodings into propositional satisfiability (SAT)

The Propositional Satisfiability Problem (SAT) Formula in propositional logic i.e. it only contains propositional (Boolean) variables Shorthand notation: x for X=true, and x for X=false Literal: x, x In so-called conjunctive normal form (CNF) Conjunction of clauses (disjunctions of literals) E.g., F = (x1  x2  x3)  (x1  x2  x3)  (x1  x2  x3) Let’s write F as a CSP: 3 variables: X1, X2, X3 Domains: for all variables {true, false} Constraints (clauses): (x1  x2  x3) (x1  x2  x3) (x1  x2  x3) One of the models: X1 = true, X2 = false, X3 = true

Importance of SAT Similar problems as in CSPs Decide whether F has a model Find a model of F First problem shown to be NP-hard problem (3-SAT) One of the most important problems in theoretical computer science Is there an efficient (i.e. worst-case polynomial) algorithm for SAT? I.e., is NP = P? SAT is a deceptively simple problem! Important in practice: encodings of formal verification problems Software verification: finding bugs in Windows etc. Hardware verification: verify computer chips (IBM, Intel big players)

SAT Solvers Building algorithms and software that perform well in practice On the type of instances they face Software and hardware verification instances 100,000s of variables, millions of constraints Runtime: seconds! But: there are classes of instances where current algorithms fail International SAT competition (http://www.satcompetition.org/) About 40 solvers from around the world compete, bi-yearly Best solver in 2007 and 2009: SATzilla: a SAT solver monster (combines many other SAT solvers) Lin Xu, Frank Hutter, Holger Hoos, and Kevin Leyton-Brown (all from UBC)

Lecture Overview Constraint Satisfaction Problems (CSPs): Definition and Recap Constraint Satisfaction Problems (CSPs): Motivation Solving Constraint Satisfaction Problems (CSPs) Generate & Test Graph search

Generate and Test (GT) Algorithms Systematically check all possible worlds Possible worlds: cross product of domains dom(V1)  dom(V2)  ...  dom(Vn) # possible worlds = |dom(V1)|  |dom(V2)|  ...  |dom(Vn)| Generate and Test: Generate possible worlds one at a time Test constraints for each one. Example: 3 variables A,B,C For a in dom(A) For b in dom(B) For c in dom(C) if {A=a, B=b, C=c} satisfies all constraints return {A=a, B=b, C=c} %This possible world is a model fail

Generate and Test (GT) Algorithms If there are k variables, each with domain size d, and there are c constraints, the (worst-case time) complexity of Generate & Test is There are dk possible worlds For each one need to check c constraints O(ckd) O(ckd) O(cdk) O(dck)

Lecture Overview Constraint Satisfaction Problems (CSPs): Definition and Recap Constraint Satisfaction Problems (CSPs): Motivation Solving Constraint Satisfaction Problems (CSPs) Generate & Test Graph search

CSP as a Search Problem: one formulation States: partial assignment of values to variables Start state: empty assignment Successor function: states with the next variable assigned E.g., follow a total order of the variables V1, …, Vn A state assigns values to the first k variables: {V1 = v1,…,Vk = vk } Neighbors of node {V1 = v1,…,Vk = vk }: nodes {V1 = v1,…,Vk = vk, Vk+1 = x} for each x  dom(Vk+1) Goal state: complete assignments of values to variables that satisfy all constraints That is, models Solution: assignment {V1 = v1,…,Vn = vn } (the path doesn’t matter)

Which search algorithm would be most appropriate for this formulation of CSP? Depth First Search Least Cost First Search A * None of the above

Relationship To Search The path to a goal isn’t important, only the solution is Heuristic function: “none” All goals are at the same depth CSP problems can be huge Thousands of variables Exponentially more search states Exhaustive search is typically infeasible Many algorithms exploit the structure provided by the goal  set of constraints, *not* black box

Backtracking algorithms Explore search space via DFS but evaluate each constraint as soon as all its variables are bound. Any partial assignment that doesn’t satisfy the constraint can be pruned. Example: 3 variables A, B, C each with domain {1,2,3,4} {A = 1, B = 1} is inconsistent with constraint A  B regardless of the value of the other variables  Fail! Prune!

CSP as Graph Searching {} V1 = v1 V1 = vk V1 = v1 V2 = v1 V2 = v2 Check unary constraints on V1 If not satisfied = PRUNE V1 = v1 V1 = vk V1 = v1 V2 = v1 V2 = v2 V2 = vk Check constraints on V1 and V2 If not satisfied = PRUNE V1 = v1 V2 = v1 V3 = v2 V3 = v1

CSP as Graph Searching Problem? Performance heavily depends {} Check unary constraints on V1 If not satisfied = PRUNE V1 = v1 V1 = vk V1 = v1 V2 = v1 V2 = v2 V2 = vk Check constraints on V1 and V2 If not satisfied = PRUNE V1 = v1 V2 = v1 V3 = v2 V3 = v1 Problem? Performance heavily depends on the order in which variables are considered. E.g. only 2 constraints: Vn=Vn-1 and Vn Vn-1

CSP as a Search Problem: another formulation States: partial assignment of values to variables Start state: empty assignment Successor function: states with the next variable assigned Assign any previously unassigned variable A state assigns values to some subset of variables: E.g. {V7 = v1, V2 = v1, V15 = v1} Neighbors of node {V7 = v1, V2 = v1, V15 = v1}: nodes {V7 = v1, V2 = v1, V15 = v1, Vx = y} for any variable Vx V \ {V7, V2, V15} and all values ydom(Vx) No fixed variable ordering for this search variant Goal state: complete assignments of values to variables that satisfy all constraints That is, models Solution: assignment (the path doesn’t matter)

CSP Solving as Graph Searching 3 Variables: A,B,C. All with domains = {1,2,3,4} Constraints: A<B, B<C

Selecting variables in a smart way Backtracking relies on one or more heuristics to select which variable to consider next. E.g, variable involved in the largest number of constraints: “If you are going to fail on this branch, fail early!” Can also be smart about which values to consider first This is a different use of the word ‘heuristic’! Still true in this context Can be computed cheaply during the search Provides guidance to the search algorithm But not true anymore in this context ‘Estimate of the distance to the goal’ Both meanings are used frequently in the AI literature. ‘heuristic’ means ‘serves to discover’: goal-oriented. Does not mean ‘unreliable’!

Standard Search vs. Specific R&R systems Constraint Satisfaction (Problems): State: assignments of values to a subset of the variables Successor function: assign values to a “free” variable Goal test: all variables assigned a value and all constraints satisfied? Solution: possible world that satisfies the constraints Heuristic function: none (all solutions at the same distance from start) Planning : State Successor function Goal test Solution Heuristic function Inference

Learning Goals for today’s class Define possible worlds in term of variables and their domains Compute number of possible worlds on real examples Specify constraints to represent real world problems differentiating between: Unary and k-ary constraints List vs. function format Verify whether a possible world satisfies a set of constraints (i.e., whether it is a model, a solution) Implement the Generate-and-Test Algorithm. Explain its disadvantages. Solve a CSP by search (specify neighbors, states, start state, goal state). Compare strategies for CSP search. Implement pruning for DFS search in a CSP. Coming up: Arc consistency and domain splitting Read Sections 4.5-4.6 Assignment 1 is due this Friday